Brain4J 2.9.1

What's New 🚀

Minor Updates

• Smart Trainer: Fixed issue where evaluation is null

Full Changelog: View changes

Brain4J 2.9.0

What's New 🚀

Major Updates

• Transformers: Now using batches.
• Optimizations: Further performance increase.
• Package refactor: Refactored package name from net.echo to org.brain4j.
• Normalization: Integrated data source normalization.
• Visuals: Improved visuals.

Features

• More TensorCPU optimizations: by @RaynLegends in #19

Full Changelog: View changes

Brain4J 2.8.0

What's New 🚀

Major Updates

• SIMD: Implemented SIMD support for better performances.
• Multi-Dimensional Ops: Implemented matrix multiplication for multi-dimensional tensors.
• Auto Encoders: Implemented auto encoders support.
• Lion Optimizer: Implemented Lion optimizer.
• Cross-Entropy: Now returns the average of the loss instead of the total value.
• Benchmarks: Added benchmarks on the documentation.

Features

• TensorCPU optimization: by @RaynLegends in #15

Full Changelog: View changes

Brain4J 2.7.1

What's New 🚀

Updates

• JitPack Deploy: Fixed jitpack deployment

Full Changelog: [View changes](2.7.0...2.7.1)

Brain4J 2.7.0

What's New 🚀

Major Updates

• Reworked Batching: Refactored batch processing for better performance and scalability.
• Tensor API: Full implementation of Tensor API for seamless tensor operations.
• Performance Boost: Significant speedup in the entire library.
• Transformer API: Improved API for transformer models with added flexibility.
• Memory Efficiency: Removed Synapse and Neuron classes in favor of the Tensor API, reducing memory usage.

Library Structure

• Project Subdivision: Split into two subprojects for modularity:
  • brain4j-core: Core functionality
  • brain4j-math: Math operations

Features

• Tensor Algebra: Added tensor algebra support by @Adversing (#5).
• Transformers: New tensor normalization and prediction methods by @Adversing (#6).

Deprecated

• API Deprecations:
  • DataSet/DataRow API
  • Vector API

Full Changelog: View changes

Brain4J 2.6.0

🚀 Updates

• Implemented almost completely Convolutional Neural Networks (CNNs)
• Implemented a better evaluation system
• Started working on Recurrent Neural Networks (RNNs)
• Refactored transformer encoder and decoder implementations
• Refactored Model class to be more flexible
• Optimized memory usage by 100%+

Full Changelog: 2.5.5...v2.6.0

Brain4J 2.5.5

Changes

• Reorganized code
• Improved wiki
• Refactored DataSet
• Fixed delta initializing bug

The training is now much better and stable.

Full Changelog: 2.5.4...2.5.5

Brain4J 2.5.4

Changes

• Fixed automatic partition bug

Full Changelog: 2.5.3...2.5.4

Brain4J 2.5.3

Changes

• Refactored activation functions for better compatibility by @Adversing in #2
• Fix: merge conflicts by @Adversing in #3
• General code refactor & Optimizations
• Removed BatchedUpdater
• Better threading management
• Convolutional Neural Networks initial work
• Renamed CATEGORICAL_CROSS_ENTROPY to CROSS_ENTROPY
• DataSet#partition is no longer mandatory
• Changed model backup format
• Fixed model not loading properly biases
• Small changes to the training listener
• Added mish and swish activations
• Updated github workflow version

Full Changelog: https://github.com/xEcho1337/brain4j/compare/2.5.2

2.5.3-pre3

Changes

• Added a "took" parameter inside the train listener
• Added mish and swish activations
• Updated github workflow version

Full Changelog: 2.5.3-pre2...2.5.3-pre3